Skip to content

UB Customizations

Source: .agents/skills/ub-customizations/SKILL.md

ub-customizations helps choose and build reusable agent customization artifacts: skills, hooks, MCP configs, prompts, bundles, and validation surfaces.

Core Principles

  • Classify the artifact before generating it.
  • Choose the smallest customization surface that solves the problem.
  • Separate reusable guidance from deterministic automation.
  • Use interviews when artifact choice materially changes the result.
  • Keep secrets and environment-specific values outside generated artifacts.
  • Validate generated skills, hooks, MCP configs, and bundles with the appropriate checklist.

Behavior In Practice

  • Classifies the request before generating files: skill for reusable on-demand guidance, hook for deterministic lifecycle automation, MCP config for real external capabilities, and bundle only when multiple artifacts materially improve the workflow.
  • Interviews only for unresolved path-changing facts: automatic lifecycle events, external systems, credentials, bundled references, host portability, and companion artifacts.
  • Treats hooks as automation, not advice. If the user needs guaranteed behavior before or after agent lifecycle events, the skill prefers a hook plus a small helper script over a long inline shell command.
  • Treats MCP as capability expansion. If the workflow needs authenticated APIs, databases, browsers, remote services, or tools outside the workspace, it designs explicit transports, inputs, names, and trust boundaries.
  • Keeps skills lean with progressive disclosure: the main SKILL.md routes and orchestrates, while references, scripts, or assets carry detail only when they are needed.
  • Applies least privilege by default: narrow tools, no hardcoded secrets, explicit environment or input variables, and warnings for preview features or third-party trust.
  • Validates generated artifacts with smoke-test prompts and portability notes so users know whether a customization is VS Code-specific, Copilot-specific, or broadly portable.

Reference Highlights

  • .agents/skills/ub-customizations/references/skills.md: skill anatomy, metadata, routing descriptions, progressive disclosure, bundled references, scripts, assets, and generation checklist.
  • .agents/skills/ub-customizations/references/hooks.md: hook events, file structure, deterministic automation examples, wrapper-script defaults, destructive-command guardrails, and preview-status warnings.
  • .agents/skills/ub-customizations/references/mcp.md: MCP server config, stdio versus HTTP transport choices, secret inputs, sandbox support, capability naming, and generation defaults.
  • .agents/skills/ub-customizations/references/bundles.md: when to combine skills, hooks, MCP configs, and helper scripts without building a monolith.
  • .agents/skills/ub-customizations/references/validation.md: artifact validation checklists, trigger-evaluation prompts, MCP and hook smoke tests, and human review questions.
  • .agents/skills/ub-customizations/references/prompt-engineering.md: instruction structure, routing-description formula, examples-over-advice, tool guidance rules, progress patterns, and naming conventions.

Progressive Disclosure

The main skill performs classification and high-level workflow. It loads the artifact-specific reference only after the artifact type is clear: skills for skill generation, hooks for lifecycle automation, MCP for external tools, and validation for final review.

Common Invocation Examples

  • “Use ub-customizations to decide whether this should be a skill or hook.”
  • “Create a skill for this reusable workflow.”
  • “Review this MCP config and companion skill design.”
  • “Package this as a bundle with validation guidance.”

Boundaries

Do not use it for general code implementation or governance-only decisions. Use ub-authoring when the work is only reusable wording, routing quality, or shared authoring convention cleanup.

Tradeoffs

Strength: prevents overbuilding customization systems and keeps artifact choice explicit.

Cost: classification can require an interview before implementation when the right artifact is not obvious.

Uncle Bob skill system documentation

Scroll to zoom · Drag to pan